Defocus-to-focus SEM Restoration on Real defocused SEM images Aggregate S→R protocol (All)
0CD (MAE)Focused Input
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Focused InputInput Type=Reference2026.04 | 0 | 0 | |
| ViT-MAE + MoE (ℓ1)Architecture=ViT-MAE + MoE, Loss=ℓ12026.04 | 0.1 | 0.52 | |
| ViT-MAE + MoE (ℓ1+TV)Architecture=ViT-MAE + MoE, Loss=ℓ1 + TV2026.04 | 0.35 | 0.53 | |
| ViT-MAE (SEM)Pre-training Dataset=SEM, Architecture=ViT-MAE2026.04 | 0.55 | 0.66 | |
| Defocused InputInput Type=Noisy/Defocused2026.04 | 0.65 | 0.91 | |
| BM3DMethod Type=Denoising2026.04 | 0.65 | 0.5 | |
| Richardson–LucyMethod Type=Classical Deconvolution2026.04 | 1.05 | 0.6 | |
| Noise2Noise (N2N)Architecture=UNet2026.04 | 1.3 | 0.7 | |
| Noise2Void (N2V)Architecture=UNet2026.04 | 1.3 | 0.7 | |
| MRNArchitecture=UNet2026.04 | 2.7 | 0.92 |